三种Uncertainty Sampling主动学习的reference

In the context of active learning.

The uncertainty instances are known as the most informative instances.

In uncertainty sampling, which is among the most popular approaches, the active learner sequentially queries the label of these instances for which its current prediction is maximally uncertain.

Uncertainty reduction methods assume that the base learner provides a reliable estimate of its confidence in each prediction.

1：maximum entropy 

F. Jing, M. Li, H. Zhang, and B. Zhang, “Entropy-based active learning with support vector machines for content-based image retrieval,” in ICME, 2004, pp. 85–88

Entropy is a well-known measure in information theory to determine the disorder of a system.
 

2：least confidence

Settles defines the most valuable sample as the one with the lowest maximum predicted probability among classe